What does brand safety mean here?
Brand safety with an AI employee means two things. First, that what it writes sounds like the brand: the voice, the vocabulary, the things the brand does and does not say. Second, that it never publishes a claim the brand cannot stand behind, whether about the product, a competitor, or an outcome.
Both risks are real because language models write plausibly by default. A plausible page in the wrong voice, or a plausible claim with no basis, is exactly what a brand should worry about, and each has a specific control on ftes.ai.
What is the Brand Brain?
The Brand Brain is a Digital FTE's memory of the brand: what the company does, who it serves, how it speaks, which words it uses and avoids, what it claims and what it never claims. It is built from what the platform already knows and from the documents you provide, every item shows its provenance, and you confirm or correct it, with a quarterly review.
A Social Growth FTE writing a post or a Content FTE drafting a page works from the Brand Brain, so voice is a property of memory rather than something re-prompted each time. When the brand changes, the Brand Brain is updated once and every role inherits it.
How does the fact gate protect the brand?
The fact gate blocks any claim in a draft that lacks a source among the approved documents or a named public source. Product claims trace to your documentation; comparative claims trace to a named source or are blocked; outcome claims trace to published data or are blocked.
This is what stops the plausible-but-indefensible claim. The FTE cannot publish that the product is the fastest in its class unless a source you approved says so, and the reviewer can see the source when it does.
- Product claims: from your documentation
- Comparative claims: named source or blocked
- Outcome claims: published data or blocked
- Regulated claims: blocked without an approved source, always gated
What goes in the remit for brand safety?
The remit carries the rules that are not facts: the tone for each channel, the topics the brand does not comment on, the audiences that are protected, such as never engaging with a named competitor's customers, and the situations that always escalate, such as any reply to a complaint or a controversy.
A Social Growth FTE, for example, drafts native posts per platform and proposed replies, but its remit can forbid replies on certain topics and route any reply to a negative comment to a person. The remit is written at hiring and tightened from the approval record.
Why does approval on everything public matter?
Approval on everything public is the control that holds when the others miss. Nothing publishes, posts, or sends in the brand's name without a named person approving it, and that gate can stay in place indefinitely for public content; the autonomy ladder is optional, and many brands never promote public publishing at all.
The audit trail records who approved each piece, so if something does go out that should not have, the question of who decided has an answer. Brand safety is the combination: memory for voice, the gate for claims, the remit for rules, and approval for everything else.
What does a brand-safety incident look like, and how is it handled?
A brand-safety incident with a Digital FTE is usually caught before it is public, because public content is gated: a draft post in the wrong register, or a reply to a heated comment that should have escalated, sits in Approvals and is sent back with a reason. The reason becomes a Brand Brain correction or a remit rule the same day, and the next draft inherits it. That is the common case, and it is why the approval gate on public content is worth keeping indefinitely.
The rarer case is a promoted class producing something that goes live and should not have. The response is the incident plan: step the class back to gated in Setup with one logged change, trace the item in the audit trail to see what it proposed, what evidence or memory it drew on, and under which grant it was auto-approved, fix the cause in the remit or the Brand Brain, correct or remove the item through the normal publishing route, and record what changed. The class earns its way back only through a fresh clean record. A brand that keeps public publishing gated has no second case; one that promotes narrowly has a fast, logged path back.